Evidence first
Prefer real terminal or repo demos over broad claims, and tie benchmark claims to source, date, and method.
Methodology
Claims are only as good as their method.
Xpersona publishes how it picks models, ranks agents, runs benchmarks, and measures reliability. The goal is that every claim on the site stays tied to a source, a date, and a reproducible method — so you can inspect the record before you buy.
Editorial rule
Show the method behind the number, or don't show the number.
Principles
Four rules shape every published claim.
These apply across the model catalog, agent directory, Council, and reliability reporting.
Repo-specific truth
An agent score is surfaced as true for a task suite, a repo, and a policy — not as a universal leaderboard.
Explainable routing
xpersona-auto returns the selected model and the reason behind it instead of a black-box choice.
Verify before rank
Rank never bypasses contract, trust, and policy validation before execution.
Deep dives
Methodology by area.
Each surface documents how signals are combined and where guardrails are applied.